摘要：This research was conducted to test the semi-strong efficiency of the preferred shares of Petrobras (PETR4)from the disclosure of its financial statements using the Artificial Neural Networks (ANN) technique as apredictor model of the normal value of securities. The choice of ANN is derived from the consistent results thatthis technique has achieved in problems of financial scope. The database used was obtained fromBM&FBovespa and the Securities and Exchange Commission of Brazil, with data covering the period fromSeptember 2, 2006, to August 17, 2013. The abnormal returns were calculated for a series of 23 Petrobrasevents identical in nature to the aforementioned events. The calculation of the expected value of PETR4 wasbased on the ANN multilayer perceptron, and for each estimation window, a period of 540 days wasestablished, while for the event windows, the strategy of analyzing the effects eight days before and after theannouncement was used. The results obtained do not allow a generic market classification regarding thePETR4 as efficient in its semi-strong form for events of the disclosure of financial results, as three of the 23events analyzed showed the market’s inefficiency to price the shares. This finding is supported by Student’s ttest.
关键词：Efficient Market; Semi-Strong Efficiency; Artificial Neural Networks; Study of Events.
